Anthropic’s $1.5 Billion AI Deal Covers Trump, Four Justices

Four sitting US Supreme Court justices and a trio of presidents are eligible to collect checks from Anthropic PBC’s $1.5 billion deal with authors whose books were part of a pirated database the AI company downloaded.

Books by Justices Clarence Thomas, Sonia Sotomayor, Elena Kagan, and Neil M. Gorsuch, and by Presidents Donald Trump, Joe Biden, and Barack Obama appear in a digital library of millions of pirated works Anthropic downloaded to develop its AI models. Skyhorse Publishing Inc., a publisher of Trump’s book “Time to Get Tough,” told Bloomberg Law it intends to file a payment claim, but it’s unclear which other officials’ publishers have or will also do so.

The largest ever copyright settlement stems from a class action claiming Anthropic illegally trained its large language models on protected works without consent. The company struck the historic deal months before trial, citing “inordinate pressure” to settle and avoid potentially business-ending damages of $1 trillion after a federal judge certified the class of authors.

Accepting settlement money could create ethics problems, especially for the jurists. Should these authors file claims, they would be eligible for payouts up to $3,100 per book, according to the database created to identify works included in the record settlement.

The legality of AI training on copyrighted works—a central question in the Anthropic case—is likely to reach the Supreme Court, creating a potential conflict of interest. Although Anthropic settled this dispute before trial, the pipeline of AI training copyright lawsuits remains robust, with dozens of cases primed to climb through the circuits and up to the high court.

Tony Lyons, president of Skyhorse Publishing, said in an emailed statement that it plans to submit a claim for Trump’s book, but it’s unclear if the publisher will evenly split the reward with the president—the default assumption under the settlement—or request a different allocation. Other publishers declined to comment or didn’t respond to questions about plans to file claims for the presidents, vice presidents, and judges.

Third Circuit Judge Stephanos Bibas is eligible to file two claims. Sitting in a federal trial court, Bibas in February ruled an AI-powered legal tool’s ingestion of protected data isn’t fair use under copyright law, and an appeal of the decision is pending. He declined to comment on whether he plans to submit claims.

Books by at least three other appellate judges, David J. Barron of the First Circuit, Senior Judge Guido Calabresi of the Second Circuit, and Senior Judge Jay S. Bybee of the Ninth Circuit, also appeared in Anthropic’s pirated library and qualify for settlement payments, according to a Bloomberg Law review.

Ethical Concerns

While federal judges are subject to enforceable ethics codes, the nine Supreme Court justices self-regulate under the federal law requiring jurists to recuse themselves from cases if their “impartiality might reasonably be questioned,” leading to sometimes uneven enforcement. The code of ethics adopted by the court in 2023 is “an aspirational code with no enforcement mechanism,” said Hofstra University law professor James Sample.

Emory University law professor Tonja Jacobi said the justices should instruct their publishers not to file a claim to avoid any perception of bias when AI training questions reach the high court. Five justices in May recused themselves from a copyright case involving a Penguin Random House subsidiary, which has published books authored by four of the justices. Because the court lacked a quorum it was required by statute to affirm the judgment of the lower court.

The justices may disagree among themselves on the propriety of collecting the settlement money. Either way, New York University law professor Stephen Gillers sees no problem with justices opting in because the AI training matter isn’t presently before them and the dollar amount is too low to merit disqualification.

“These are not poor people,” he said, “so it’s not as though even a low six-figure sum"—let alone $3,100 per title—"will affect their quality of life in any way.”

Bubbling Issue

Authors, photographers, and news publishers have filed dozens of copyright lawsuits against leaders of the burgeoning AI industry since 2022, largely claiming the companies illegally trained large language models on copyrighted works without permission or compensation. While the Anthropic settlement centers on pirate downloads, the novel question of whether the training qualifies as fair use is slowly ascending the legal ladder.

Two federal California judges this year issued divergent opinions. Judge William Alsup, who oversaw the case Anthropic settled, found the training to be fair use, but downloading pirated books likely wasn’t. Judge Vince Chhabria also found Meta Platforms Inc.'s training was fair use, but indicated he would’ve ruled for the plaintiffs if they’d presented better evidence of market harm.

The case closest to the Supreme Court is Thomson Reuters’ lawsuit against the AI-powered legal tool Ross Intelligence Inc. over its ingestion of legal headnotes. The US Court of Appeals for the Third Circuit in June accepted Ross’ early appeal of a summary judgment ruling rejecting its fair use defense.

There’s “a lot more pressure” on lower court judges to recuse, Jacobi said, because their code of ethics has a real enforcement mechanism.

Supreme Court justices have also argued lower court judges can more easily recuse themselves, because there are more of them.

That’s another reason they should refuse any settlement money, Jacobi said.

“If they’re publishing books that are contributing to the understanding of the law in the general community, which they should be, they shouldn’t be focused on how much money they can make,” she said.